yichen-xiaohongshu-fetch
Pass
Audited by Gen Agent Trust Hub on Jul 20, 2026
Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructions and scripts involve executing Python scripts for fetching and transcribing data.
- Evidence:
python3 yichen-xiaohongshu-fetch/scripts/fetch.py "<小红书笔记链接>" [输出目录]inSKILL.md. - Evidence:
python3 yichen-volc-asr/scripts/transcribe.py '<视频文件路径>'inSKILL.md. - [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted content from Xiaohongshu (transcripts and OCR text) which is then processed by the LLM to generate summaries and JSON structures.
- Ingestion points: Scraped HTML, video transcripts (ASR), and OCR text from images fetched by
fetch.py. - Boundary markers: Absent. The prompt instructions in Step 4 of the video workflow and Step 2/3 of the image workflow do not include delimiters or safety warnings for the untrusted data.
- Capability inventory: The agent can execute shell commands (
python3,curl), write to the filesystem, and perform network requests to the Feishu API. - Sanitization: No sanitization or filtering of the external content is mentioned before it is passed to the LLM.
Audit Metadata